Base station, apparatus, method, and program for hybrid numerology configuration selection

Abstract

The requirement feature extractor (21) extracts statistical features related to the service requirements of the UE. The traffic feature extractor (22) extracts statistical features related to transmission and reception traffic. The channel feature extractor (23) extracts statistical features related to radio channel conditions and radio channel configurations. The context unit (24) generates a context vector based on the statistical features. The ML model array (25) estimates the QoS of the mixed numerology configuration based on the context vector. The decision unit (26) selects the mixed numerology configuration to be used for data transmission and reception based on the estimated QoS.

Description

Technical Field

[0001] The present disclosure relates to a base station, an apparatus, a method, and a computer-readable medium for hybrid numerology configuration selection.

Background Art

[0002] Hybrid numerology communication has been proposed because fixed numerology systems lack sufficient flexibility to accommodate the diverse requirements (e.g., latency, error rate, etc.) of a large number of devices that may need to share the spectrum. For example, Internet of Things (IoT) devices such as sensors and autonomous vehicles have completely different requirements regarding packet error tolerance and latency, and thus cannot be managed in a fixed numerology system. When devices with different requirements use the same numerology, the quality of service (QoS) experienced by end users may degrade.

[0003] To improve network flexibility, methods such as the greedy method have been proposed (Non-Patent Document 1). Furthermore, a low computational complexity method using a lookup table approach has been proposed to provide a feasible set of hybrid numerology configurations (Patent Document 1). Other approaches such as Doppler spread estimation for selecting a hybrid numerology configuration have been proposed (Patent Document 2). Each associated user equipment (UE) separately transmits movement information using on-board sensors.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Non-Patent Documents

[0005] [Non-Patent Document 1] A. Yazar and H. Arslan, "A Flexibility Metric and Optimization Methods for Mixed Numerologies in 5G and Beyond," in IEEE Access, vol. 6, pp. 3755-3764, 2018, doi: 10.1109 / ACCESS.2018.2795752. [Summary of the Invention] [Problems to be Solved by the Invention]

[0006] Despite the method disclosed in Patent Document 1, this method reduces the search space for possible mixed numerology configurations. This is useful, but Patent Document 1 does not provide a method for identifying the optimal mixed numerology configuration within the set of executable ones.

[0007] The mobile information method disclosed in Patent Document 2 mitigates the adverse effects of Doppler spread. However, this method requires the periodic transmission of mobile information from each UE to the next-generation NodeB (gNB), thereby imposing a large overhead on the network. Furthermore, Patent Document 2 does not consider the service requirements of non-mobile UEs.

[0008] The greedy method disclosed in Non-Patent Document 1 uses brute-force calculation to calculate the network flexibility of all possible mixed numerology configurations. This is theoretically possible, but it is not actually scalable because the computational complexity increases exponentially with the number of UEs. Therefore, this method is not suitable for high-density wireless networks.

[0009] Therefore, an example of the object of the present disclosure is to achieve high QoS for a wireless network having UEs with diverse service requirements without adding additional overhead.

Means for Solving the Problem

[0010] To achieve the above-exemplified object, in a first aspect, the present disclosure provides an apparatus for hybrid numerology configuration selection. The apparatus includes a requirement feature extractor configured to extract statistical features related to service requirements of UEs associated with a base station, a traffic feature extractor configured to extract statistical features related to transmission / reception traffic of a base station, a channel feature extractor configured to extract statistical features related to radio channel conditions and radio channel configurations, a context unit configured to generate a context vector based on the statistical features related to service requirements, the statistical features related to the transmission / reception traffic, and the statistical features related to the radio channel conditions and radio channel configurations, a first ML model array including a plurality of machine learning (ML) models each configured to estimate the quality of service (QoS) of a hybrid numerology configuration based on the context vector generated by the context unit, and a determination unit configured to select a hybrid numerology configuration used for data transmission / reception based on the QoS estimated by the ML model.

[0011] In a second aspect, the present disclosure provides a base station. The base station includes a communication unit configured to communicate with a core network and a UE, and a hybrid numerology configuration selection unit configured to select a hybrid numerology configuration used for data transmission / reception. The hybrid numerology configuration selection unit includes a requirement feature extractor configured to extract statistical features related to service requirements of the UE, a traffic feature extractor configured to extract statistical features related to transmission / reception traffic of the base station, A channel feature extractor configured to extract statistical features related to wireless channel conditions and wireless channel configurations, A context unit configured to generate a context vector based on the statistical features related to the service requirements, the statistical features related to the transmission and reception traffic, and the statistical features related to the wireless channel conditions and wireless channel configurations, A first ML model array including a plurality of machine learning (ML) models each configured to estimate the quality of service (QoS) of the mixed numerology configuration based on the context vector generated by the context unit, And a determination unit configured to select a mixed numerology configuration used for data transmission and reception based on the QoS estimated by the ML model.

[0012] In a third aspect, the present disclosure provides a method for mixed numerology configuration selection. The method includes: Extracting statistical features related to the service requirements of UEs associated with a base station, Extracting statistical features related to the transmission and reception traffic of the base station, Extracting statistical features related to wireless channel conditions and wireless channel configurations, Generating a context vector based on the statistical features related to the service requirements, the statistical features related to the transmission and reception traffic, and the statistical features related to the wireless channel conditions and wireless channel configurations, Using a machine learning (ML) model array including a plurality of ML models each configured to estimate the quality of service (QoS) of the mixed numerology configuration, and estimating the QoS of the mixed numerology configuration based on the context vector, And selecting a mixed numerology configuration used for data transmission and data reception based on the QoS estimated by the ML model.

[0013] This disclosure provides a non - transitory computer - readable medium for hybrid numerology configuration selection in a fourth disclosure. The non - transitory computer - readable medium extracts statistical features related to the service requirements of UEs associated with a base station, extracts statistical features related to the transmission and reception traffic of the base station, extracts statistical features related to radio channel conditions and radio channel configurations, generates a context vector based on the statistical features related to service requirements, the statistical features related to transmission and reception traffic, and the statistical features related to radio channel conditions and radio channel configurations, uses a machine learning (ML) model array including a plurality of ML models each configured to estimate the QoS of a hybrid numerology configuration, and estimates the quality of service (QoS) of the hybrid numerology configuration based on the context vector, stores a program for causing a computer to select a hybrid numerology configuration used for data transmission and data reception based on the QoS estimated by the ML model.

Advantages of the Invention

[0014] A base station, an apparatus, a method, and a computer - readable medium for hybrid numerology configuration selection can achieve high QoS for a wireless network having UEs with diverse service requirements without adding additional overhead.

Brief Description of the Drawings

[0015]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7A

Figure 7B

Figure 8

Figure 9

Figure 10

Mode for Carrying Out the Invention

[0016] <Overview> Before describing the embodiments of the present disclosure, an overview of the present disclosure will be described. Figure 1 shows an example of the schematic configuration of a base station according to the present disclosure. The base station 10 includes a communication unit 15 and a mixed numerology configuration selection unit 20. The communication unit 15 communicates with the core network and the UE. The mixed numerology configuration selection unit 20 is a device for selecting a mixed numerology configuration. The mixed numerology configuration selection unit 20 includes a requirement feature extractor 21, a traffic feature extractor 22, a channel feature extractor 23, a context unit 24, a machine learning (ML) model array 25, and a decision unit 26.

[0017] The requirement feature extractor 21 extracts statistical features related to the service requirements of the UEs associated with the base station 10. The traffic feature extractor 22 extracts statistical features related to the transmission and reception traffic of the base station 10. The channel feature extractor 23 extracts statistical features related to the radio channel conditions and radio channel configurations. The context unit 24 generates a context vector based on the statistical features related to the service requirements, the statistical features related to the transmission and reception traffic, and the statistical features related to the radio channel conditions and radio channel configurations.

[0018] The ML model array 25 includes a plurality of ML models 30. Each of the ML models 30 is configured to estimate the QoS of the hybrid numerology configuration based on the context vector generated by the context unit 24. The decision unit 26 selects the hybrid numerology configuration to be used for data transmission and reception based on the QoS estimated by the ML model 30.

[0019] According to the present disclosure, the requirement feature extractor 21, the traffic feature extractor 22, and the channel feature extractor 23 extract statistical features related to the service requirements, the transmission and reception traffic, and the radio channel conditions and radio channel configurations. The context unit 24 generates a context vector based on the above statistical features. The ML model array 25 uses the context vector to QoS estimate. In the present disclosure, the ML model array 25 can estimate the QoS of the hybrid numerology configuration according to the statistical features of the wireless communication between the base station 10 and the UE. The decision unit 26 uses the QoS estimated by the ML model 30 to select the hybrid numerology configuration. Thereby, the present disclosure can achieve high QoS for wireless communication with UEs having various service requirements without adding additional overhead.

[0020] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the embodiments, the same or similar elements are denoted by the same reference numerals, and redundant descriptions are omitted.

[0021] <First Embodiment> FIG. 2 shows a wireless communication system including a base station according to the first embodiment of the present disclosure. The communication system 100 includes a 5G core network 110, a UE 120, and a base station 300. It is assumed that the base station 300 is a base station in a fifth-generation (5G) wireless communication network. The 5G core network (5GC) 110 is a mobile core network system that accommodates 5G wireless. The UE 120 can communicate with the 5GC 110 via the base station 300. The UE 120 may have various service requirements. The communication system 100 is configured as a mixed numerology system for accommodating various requirements of the UE 120. A mixed numerology system is known as a system that enables simultaneous use of multiple numerologies within an available bandwidth.

[0022] FIG. 3 shows a table of numerologies supported in different frequency bands according to the 5G NR (New Radio) Release 15 specification of the ETSI (European Telecommunications Standards Institute). Each numerology is identified by a parameter μ and is associated with different channel parameters such as carrier spacing, cyclic prefix period, slot period, etc. FIG. 3 is for illustrative purposes and is not limited by numerologies to be supported in the future.

[0023] FIG. 4 shows an example of a bandwidth. As shown in FIG. 4, in the data frame 1 201-1, the bandwidth 202 is divided into different numerology bands 204-1 to 204-3 and guard bands (GB) 203-1 to 203-6. The ordered tuple of numerology values μ in the numerology bands 204-1 to 204-3 is defined as a mixed numerology value configuration. In the example shown in FIG. 4, in the data frame 1 201-1, the mixed numerology value configuration is (0, 2, 1). In the data frame 2The hybrid numerology value configuration in 201-2 is (0). Note that the numerology bands 204-1 to 204-3 can be collectively referred to as the numerology band 204 when there is no particular need to distinguish them. In this embodiment, the values of the guard bands 203-1 to 203-6 are determined by the base station 300 (see Fig. 2), and it does not matter how it is realized. The values of the guard bands can follow any of the conventional ones. The hybrid numerology configuration does not change between data frames and can be changed at the end of any data frame.

[0024] According to this embodiment, the hybrid numerology configuration is determined by the base station 300 after the end of each data frame shown in Fig. 4. One of the purposes of this embodiment is to determine the hybrid numerology configuration for each data frame so that the overall QoS is maximized.

[0025] Fig. 5 shows a schematic configuration of the base station 300 according to this embodiment. The base station (gNB) 300 includes a Service Data Adaptation Protocol (SDAP) layer module 321, a Packet Data Convergence Protocol (PDCP) layer module 322, a Radio Link Control (RLC) layer module 323, a Medium Access Control (MAC) layer module 324, a Physical (PHY) layer module 325, a set of antennas (antenna array) 326, and a hybrid numerology selection unit 302. The gNB 300 can be configured as an Orthogonal Frequency Division Multiple Access (OFDMA) base station. The gNB 300 corresponds to the base station 10 shown in Fig. 1.

[0026] The SDAP layer module 321 transmits data to the PHY layer module 325 via the PDCP layer module 322, RLC layer module 323, and MAC layer module 324 when sending packets. The SDAP layer module 321 receives data in the reverse order when receiving data. The SDAP layer module 321 marks the QoS flow identifier (QFI) in both downlink and uplink packets, among other things. The SDAP layer module 321 periodically sends a "start" interrupt signal to the mixed numerology selection unit 302, and then sends a "stop" interrupt signal, and the gNB 300 receives the mixed numerology configuration for which data is to be transmitted and received. The time difference between the start interrupt signal and the stop interrupt signal can be considered a system hyperparameter determined by the network operator / engineer. The operation of the mixed numerology selection unit 302 will be described later. The PHY layer module 325 shares a channel quality indicator (CQI) report with the mixed numerology selection unit 302, among other things. The SDAP layer module 321, PDCP layer module 322, RLC layer module 323, MAC layer module 324, and PHY layer module 325 correspond to the communication unit 15 shown in FIG. 1.

[0027] According to this embodiment, the mixed numerology selection unit 302 includes a storage buffer 303, a requirement feature extractor 304A, a traffic feature extractor 304B, a channel feature extractor 304C, a context unit 305, a machine learning (ML) model array 306, a decision unit 307, a performance monitoring unit 308, a database 309, and a 5G QoS identifier (5QI) lookup table 310. The mixed numerology selection unit 302 corresponds to the mixed numerology configuration selection unit 20 shown in FIG. 1.

[0028] The requirement feature extractor 304A extracts statistical features related to the service requirements of the UE 120 associated with the gNB 300. For example, the requirement feature extractor 304A collects service requirements over a period of time. The service requirements include, for example, at least one of a delay budget and a packet error rate tolerance. The requirement feature extractor 304A calculates at least one statistical value of the collected service requirements as a statistical feature related to the service requirements. The requirement feature extractor 304A corresponds to the requirement feature extractor 21 shown in FIG. 1.

[0029] The traffic feature extractor 304B extracts statistical features related to the overall transmission and reception traffic. For example, the traffic feature extractor 304B collects traffic feature information of data packets over a period of time. The traffic feature information includes, for example, at least one of a UE ID (Identifier), a packet size, and a packet arrival timestamp. The traffic feature extractor 304B calculates at least one statistical value based on the collected traffic feature information as a statistical feature related to the overall transmission and reception traffic. The traffic feature extractor 304B corresponds to the traffic feature extractor 22 shown in FIG. 1.

[0030] The channel feature extractor 304C extracts statistical features related to the radio channel conditions and the radio channel configuration. For example, the channel feature extractor 304C collects channel feature information over a period of time. The channel feature information includes, for example, at least one of an available latest CQI report, the number of available resource blocks, a guard band value, a downlink and uplink share rate, and a HARQ (Hybrid Automatic Repeat reQuest) process ID. The channel feature extractor 304C calculates at least one statistical value based on the collected channel feature information as a statistical feature related to the radio channel conditions and the radio channel configuration. The channel feature extractor 304C corresponds to the channel feature extractor 23 shown in FIG. 1.

[0031] The context unit 305 generates a context vector based on inputs from the requirement feature extractor 304A, the traffic feature extractor 304B, and the channel feature extractor 304C. The context vector may be an array in which statistical features related to service requirements, statistical features related to overall transmission and reception traffic, and statistical features related to radio channel conditions and radio channel configurations are concatenated. The context unit 305 corresponds to the context unit 24 shown in FIG. 1.

[0032] The ML model array (the first ML model array) 306 is used to estimate the QoS of the mixed numerology configuration. The QoS may be the ratio of the number of QoS flows corresponding to UEs for which all service requirements are satisfied to the total number of QoS flows. The ML model array 306 includes M ML models (ML model 1 to ML model M) 306-1 to 306-M. The number (M) of ML models may be the same as the number of possible mixed numerology configurations. The ML models 306-1 to 306-M are each configured to estimate the QoS of the mixed numerology configurations 1 to M. The ML models 306-1 to 306-M are assumed to be pre-trained in an offline manner using either actual communication data or simulated communication data (synthetic data) generated by a communication network simulation. Note that the ML models 306-1 to 306-M may be generically referred to as the ML model array 306 when there is no need to distinguish them. The ML model array 306 and the ML models 306-1 to 306-M respectively correspond to the ML model array 25 and the ML models 30 shown in FIG. 1.

[0033] The decision unit 307 selects a hybrid numerology configuration to be used for data transmission and reception based on the QoS estimated by the ML model array 306. The decision unit 307 can select the hybrid numerology configuration having the highest estimated QoS among the estimation results of the ML model array 306. When the difference between the QoS estimated for the current hybrid numerology configuration and the highest estimated QoS is greater than a predetermined threshold, the decision unit 307 can select the highest estimated hybrid numerology configuration. When the difference between the QoS estimated for the current hybrid numerology configuration and the highest estimated QoS is smaller than the threshold, the decision unit 307 can determine not to change the current hybrid numerology configuration. The decision unit 307 corresponds to the decision unit 26 shown in FIG. 1.

[0034] The performance monitoring unit 308 calculates the actual QoS obtained for the hybrid numerology configuration selected by the decision unit 307. The database 309 stores the labeled context information. The labeled context information can include a context vector, the estimated optimal hybrid numerology configuration, and the corresponding actual QoS. For example, the labeled context information can include a tuple of {context vector, selected hybrid numerology configuration, obtained actual QoS}, and the selected hybrid numerology configuration and the obtained actual QoS are called labels. The labeled context information is stored for future use (e.g., offline training of at least one of the ML models 306-1 to 306-M and performance analysis of the hybrid numerology selection unit 302). For example, the database may be a fixed-size buffer, and when its capacity is reached, the oldest labeled context information may be deleted to accommodate new labeled context information.

[0035] The storage buffer 303 is configured to store the partially processed information to be processed by the requirement feature extractor 304A, the traffic feature extractor 304B, and the channel feature extractor 304C. The 5QI look-up table 310 includes service requirements corresponding to each of the 5QI values (values indicating QoS). FIG. 6 shows an example of the 5QI look-up table 310. The 5QI look-up table 310 maps 5QI values to service requirements. The service requirements include, for example, a delay budget and a packet error rate tolerance range. The requirement feature extractor 304A refers to the 5QI look-up table 310 and extracts service requirements.

[0036] The operation of the mixed numerology selector 302 (mixed numerology configuration selection method) can be divided into the following four phases. 1. Context identification phase 2. QoS estimation phase 3. Mixed numerology setting update phase 4. Performance monitoring phase

[0037] The context identification phase will be described. FIG. 7A shows the operation procedure of the context identification phase. The mixed numerology selector 302 determines whether it has received a start interrupt from the SDAP layer module 321 of the gNB 300 (step A1). When the mixed numerology selector 302 determines that it has received a start interrupt, the context identification phase is started. When the mixed numerology selector 302 does not receive a start interrupt, it waits for a start interrupt signal to be received.

[0038] In the context identification phase, the requirement feature extractor 304A extracts the service requirements of all QoS flows (step A2). In step A2, the requirement feature extractor 304A extracts, for example, the 5QI value from the SDAP header of each data packet. The requirement feature extractor 304A refers to the 5QI lookup table 310 and obtains the service requirements corresponding to the extracted 5QI value. The service requirements may include a delay budget and a packet error rate tolerance range. The requirement feature extractor temporarily stores the service requirements in the storage buffer 303 (step A3).

[0039] The traffic feature extractor 304B extracts traffic feature information (step A4). The traffic feature information may include a UE ID, a packet size, and a packet arrival timestamp. The traffic feature extractor 304B temporarily stores the extracted information in the storage buffer 303 (step A5). The channel feature extractor 304C extracts channel feature information (step A6). The channel feature information may include the latest available CQI report, the number of available resource blocks, the guard band value, the downlink and uplink share rates, and the HARQ process ID. The channel feature extractor 304C temporarily stores the extracted information in the storage buffer 303 (step A7).

[0040] The mixed numerology selector 302 determines whether it has received a stop interrupt signal from the SDAP layer module 321 (step A8). Steps A2 to A7 are repeated until it is determined in step A8 that a stop interrupt signal has been received. That is, the requirement feature extractor 304A collects and stores the service requirements in steps A2 and A3 until a stop interrupt signal is received. The traffic feature extractor 304B collects and stores the traffic feature information in steps A4 and A5 until a stop interrupt signal is received. The channel feature extractor 304C collects and stores the traffic feature information in steps A6 and A7 until a stop interrupt signal is received.

[0041] When it is determined that a stop interrupt signal has been received in step A8, the requirement feature extractor 304A calculates statistical features from the stored service requirements (step A9). The statistical features include, but are not limited to, the average value, minimum value, maximum value, median value, and variance of the delay budget and error rate tolerance range. The traffic feature extractor 304B calculates statistical features of the data traffic (step A10). The statistical features include, but are not limited to, the number of UEs, the average, minimum, maximum, median value, and variance of the packet size, the packet arrival rate, and the antenna power corresponding to the UEs. The channel feature extractor 304C calculates statistical features of the radio channel (step A11). The statistical features include, but are not limited to, the total number of resource blocks, the average, minimum, maximum, median value, and variance of the CQI, the guard band size, the guard symbol, and the downlink and uplink share rates.

[0042] The context unit 305 generates a context vector (step A12). For example, in step A12, the context unit 305 concatenates the features obtained from the requirement feature extractor 304A, the traffic feature extractor 304B, and the channel feature extractor 304C to generate a context vector. The context unit 305 stores the generated context vector in the database 309 together with a time stamp (step A13).

[0043] Next, the QoS estimation phase, the mixed numerology configuration update phase, and the performance monitoring phase will be described. FIG. 7B shows the operation procedures of the QoS estimation phase, the mixed numerology configuration update phase, and the performance monitoring phase. In the QoS estimation phase, the context unit 305 inputs the context vector to each of the ML models 306-1 to 306-M included in the ML model array 306 (step B1). The ML models 306-1 to 306-M predict or estimate the QoS for all possible mixed numerology configurations (steps B2-1 to B2-M). Each of the ML models 306-1 to 306-M outputs the estimated QoS for the corresponding mixed numerology configuration.

[0044] In the hybrid numerology configuration update phase, the determination unit 307 selects an optimal hybrid numerology configuration from possible hybrid numerology configurations based on the estimated QoS (step B3). In step B3, the determination unit 307 may select the hybrid numerology configuration having the highest QoS as the optimal hybrid numerology configuration. The determination unit 307 calculates a difference Δ between the estimated QoS in the current hybrid numerology configuration and the optimal hybrid numerology configuration (step B4). In step B4, the determination unit 307 may calculate the difference between the estimated QoS in the current hybrid numerology configuration and the highest QoS among the estimation results from step B2-1 to B2-M.

[0045] The determination unit 307 determines whether the difference Δ calculated in step B4 is greater than the hyperparameter T 1 (step B5). The hyperparameter T 1 is a threshold value of the difference Δ. If it is determined in step B5 that the difference Δ is greater than the hyperparameter T 1 , the determination unit 307 selects the optimal hybrid numerology configuration as the hybrid numerology configuration used in the SDAP layer module 321. In this case, the determination unit 307 notifies the SDAP layer module 321 of the optimal hybrid numerology configuration and updates the hybrid numerology configuration of the SDAP layer module 321 (step B6).

[0046] In the performance monitoring phase, after data transmission, the performance monitoring unit 308 calculates the actual QoS obtained for the updated hybrid numerology configuration (step B7). The performance monitoring unit 308 may calculate the ratio of the services for which the service requirements are satisfied to the total number of services as the actual QoS. The performance monitoring unit 308 stores the calculated actual QoS together with the updated hybrid numerology configuration in the database 309. If the difference Δ in step B5 is greater than the hyperparameter T 1If it is determined that it is not larger, the hybrid numerology configuration of the SDAP layer module 321 is not changed. In this case, the performance monitoring unit 308 calculates the actual QoS of the current hybrid numerology configuration at step B7 and stores the calculated actual QoS together with the current hybrid numerology configuration in the database 309. The hybrid numerology configuration, the actual QoS, and the context vector stored in the database 309 can be used for at least one of the offline training of the ML model array 306 and the performance analysis of the hybrid numerology selection unit 302.

[0047] According to this embodiment, the requirement feature extractor 304A, the traffic feature extractor 304B, and the channel feature extractor 304C extract statistical features related to service requirements, transmission and reception traffic, and radio channel conditions and radio channel configurations. The context unit 305 generates a context vector based on the above statistical features. The ML model array 306 uses the context vector to QoS estimate. In this embodiment, the ML model array 306 can estimate the QoS of the hybrid numerology configuration according to the statistical features of the wireless communication between the base station 300 and the UE 120. The decision unit 307 uses the QoS estimated by the ML model array 306 to select a hybrid numerology configuration. By doing so, the hybrid numerology selection unit 302 can select an optimal hybrid numerology configuration according to the service requirement characteristics, traffic characteristics, and channel characteristics that may change over time.

[0048] In this embodiment, the gNB 300 can autonomously and dynamically adapt the mixed numerology configuration according to the changing radio network, thereby enabling high QoS in wireless communication with the UE 120 with diverse service requirements. Also, the gNB 300 according to this embodiment can provide a medium access method that realizes a high overall throughput in a wireless communication system where a plurality of UEs can access the medium simultaneously and completely dispersedly. Furthermore, no additional overhead is required for mixed numerology selection, and the gNB 300 ensures compatibility with an 802.11 standard wireless station having a fair medium access opportunity for everyone.

[0049] <Second Embodiment Example> Next, a second embodiment example of the present disclosure will be described. FIG. 8 shows a wireless communication system including an apparatus for mixed numerology selection. The wireless communication system according to this embodiment includes a gNB 600 and an ML model update unit 650. For example, the ML model update unit 650 is included in a Radio Access Network (RAN) Intelligent Controller (RIC) that controls a base station apparatus. Note that in FIG. 8, the 5GC 110 and the UE 120 shown in FIG. 2 are omitted.

[0050] gNB 600 includes an Open RAN (O-RAN) Radio Unit (O-RU) 601, an O-RAN Distributed Unit (O-DU) 602, an O-RAN Central Unit (O-CU) 603, and a Numerology Selector Unit (NSU) 302. The O-RU 601, O-DU 602, and O-CU 603 include an SDAP layer module 321, a PDCP layer module 322, an RLC layer module 323, a MAC layer module 324, a PHY layer module 325, and an antenna array 326, as shown in FIG. 5. The configurations of the O-RU 601, O-DU 602, and O-CU 603 may be the same as those of the SDAP layer module 321, PDCP layer module 322, RLC layer module 323, MAC layer module 324, PHY layer module 325, and antenna array 326 in the first embodiment. Also, the configuration and operation of the NSU 302 may be the same as those described in the first embodiment. In this embodiment, the NSU 302 and the ML model update unit 650 constitute an apparatus for selecting a mixed numerology configuration.

[0051] In this embodiment, the ML model update unit 650 includes a control unit 651, a context buffer 652, and a pair of ML model arrays 653 and 654. The control unit 651 acquires the labeled context information stored in the database 309 (see FIG. 5) of the NSU 302. The labeled context information may include a context vector generated by the context unit 305, a mixed numerology configuration selected by the determination unit 307, and QoS calculated by the performance monitoring unit 308. The context buffer 652 is a buffer with a fixed size and is configured to store the labeled context information. The context buffer 652 may be a First In First Out (FIFO) buffer. Note that the NSU 302 and the ML model update unit 650 do not necessarily have to be configured as separate devices, and these may be included in the same device.

[0052] The ML model arrays (second model array) 653 and the ML model array (third ML model array) 654 are each used to estimate the QoS of the hybrid numerology configuration. The ML model arrays 653 and 654 are configured to replicate the operation of the ML model array 306 (see FIG. 5) in the NSU302. The ML model array 653 includes M ML models (ML model 1 to ML model M) 653-1 to 653-M. The ML model array 654 includes M ML models (ML model 1 to ML model M) 654-1 to 654-M. The control unit 651 uses the labeled context information stored in the context buffer 652 and the pair of ML model arrays 653 and 654 to update the parameters of the ML model array 306 in the NSU302 according to the radio network conditions. For example, the control unit 651 updates the parameters of the ML model array 654 using the labeled context information. The control unit 651 compares the estimation accuracy of the ML model array 653 with the estimation accuracy of the updated ML model array 654. When the estimation accuracy of the ML model array 654 is higher than the estimation accuracy of the ML model array 653, the control unit 651 updates the parameters of the ML model array 306 in the NSU302 with the parameters of the updated ML model array 654.

[0053] The operation procedure of the NSU302 may be the same as that described with reference to FIGS. 7A and 7B in the first embodiment. FIG. 9 shows the operation procedure of the ML model update unit 650. The control unit 651 periodically acquires the labeled context information from the database 309 of the NSU302 (step C1). In other words, the control unit 651 collects the past labeled context information from the database 309. In step C1, the control unit 651 can collect the labeled context information from the database 309 via the standard E2 interface of Open RAN (O-RAN). The periodicity of data acquisition is considered a hyperparameter and can be set according to the dynamic nature of the radio network. The control unit 651 stores the labeled context information in the context buffer 652.

[0054] The control unit 651 randomly samples the labeled context information to generate two mini - batches of the labeled context information (steps C2 and C3). In step C2, the control unit generates a test mini - batch including a tuple of randomly sampled labeled context information, and in step C3, the control unit generates a training mini - batch including a tuple of randomly sampled labeled context information. Each of the test mini - batch and the training test mini - batch is a randomly sampled subset of the labeled context information stored in the context buffer 652.

[0055] The control unit 651 uses the training mini - batch sampled in step C3 to update the parameters of the ML model array 654 (step C4). For example, in step C4, the control unit 651 selects one by one the labeled context information, that is, the tuples, in the training mini - batch. The control unit 651 selects an ML model for predicting QoS in the ML model array 654 according to the selected hybrid numerology configuration included in the selected tuple. The control unit 651 inputs the context vector included in the selected tuple into the selected ML model. The control unit 651 calculates the error of the selected hybrid numerology configuration based on the difference between the actual QoS included in the selected tuple and the QoS predicted by the selected ML model. The control unit 651 updates the parameters of the ML model array 654 based on the total error of the hybrid numerology configuration. The total error can be mathematically described as follows. Total error = Σ(actual QoS - predicted QoS) 2 Here, the sum is performed for all samples in the training mini - batch. Note that the total error belongs to the entire array rather than individual ML models. The parameters of the ML model array 654 are updated so that the total error is minimized.

[0056] Note that the method for updating the parameters of the ML model array 654 is not particularly limited to the above method. Any appropriate method for updating the parameters of the ML model array 654 may be used. The algorithm used to update the parameters of the ML model depends on the type of the ML model. For example, when the ML model is an ML model based on a neural network, a stochastic gradient descent algorithm may be used to update the parameters of the ML model array 654.

[0057] The control unit 651 supplies the test mini-batch sampled in step C2 to the ML model array 653 and the ML model array 654. The ML model array 653 estimates the QoS for the test mini-batch (step C5). The ML model array 654 whose parameters are updated in step C4 estimates the QoS for the test mini-batch (step C6).

[0058] The control unit 651 calculates the error difference Δ between the estimation result of the ML model array 653 and the estimation result of the ML model array 654 (step C7). In step C7, the control unit 651 may calculate the total error for each of the ML model array 653 and the ML model array 654 in the same manner as described above. The control unit 651 calculates the difference between the total error of the ML model array 653 and the total error of the ML model array 654 as the error difference Δ. The error difference Δ can be described as follows. Error difference Δ = |Total error of ML model array 653| - |Total error of ML model array 654| When the absolute value of the total error of the ML model array 654 is smaller than the absolute value of the total error of the ML model array 653, that is, when the estimation accuracy of the ML model array 654 is higher than the estimation accuracy of the ML model array 653, the error difference Δ indicates a positive value.

[0059] The control unit determines whether the error difference Δ calculated in step C7 is greater than the hyperparameter T which is the threshold value of the error difference Δ 2 (step C8). The hyperparameter T 2can be a positive value. If it is determined in step C8 that the error difference Δ is greater than the hyperparameter T 2 the control unit 651 transmits the parameters of the ML model array 654 to the NSU302 and updates the parameters of the ML model array 306 of the NSU302 (step C9). Further, the control unit 651 copies the parameters of the updated ML model array 654 to the ML model array 653 (step C10). If it is determined in step C8 that the error is not greater than the hyperparameter T 2 the parameters of the ML model array 306 of the NSU302 are not updated.

[0060] According to this embodiment, the ML model update unit 650 updates the ML model array 306 included in the NSU302 by using the labeled context information and the pair of ML model arrays 653 and 654. In this embodiment, the parameters of the ML model array 306 included in the NSU302 are updated by the ML model update unit 650 so that the error becomes smaller, thereby improving the estimation accuracy of QoS in the NSU302. Other effects are the same as those in the first embodiment.

[0061] In the above embodiment, the NSU302 and the ML model update unit 650 can be implemented by a device having one or more processors. FIG. 10 shows a configuration example of a computer device. The computer device 500 includes at least one processor 510 and at least one memory 520. The memory 520 includes at least one of a volatile memory and a non-volatile memory. The memory 520 stores software (program) executed on the processor 510, for example, in a non-volatile memory. The processor 510 is, for example, a central processing unit (CPU). The functions of the NSU302 and the ML model update unit 650 can be realized by the processor 510 operating according to a program loaded from the memory 520. The processor 510 may load a program from an external memory of the device 500.

[0062] When the above program is loaded into a computer, it includes a set of instructions (or software code) that cause the computer to execute one or more functions described in the embodiments. The program may be stored in a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, the computer-readable medium or tangible storage medium includes random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD), or other memory technologies, Compact Disc (CD), digital versatile disc (DVD), Blu-ray (registered trademark) disc, or other optical disc storage, magnetic cassette, magnetic tape, magnetic disk storage, or other magnetic storage devices. The program may also be transmitted on a transient computer-readable medium or a communication medium. By way of example and not limitation, the transient computer-readable medium or communication medium includes electrical, optical, acoustic, or other forms of propagated signals.

[0063] Although the present disclosure has been described with reference to the embodiments, the present disclosure is not limited to the above. Various changes that can be understood by those skilled in the art within the scope of the present disclosure can be added to the configuration and details of the present disclosure.

[0064] For example, all or part of the above embodiments can be described as the following supplementary notes, but are not limited thereto.

[0065] [Supplementary Note 1] An apparatus for hybrid numerology configuration selection, comprising: a requirement feature extractor configured to extract statistical features related to the service requirements of UEs associated with a base station; a traffic feature extractor configured to extract statistical features related to the transmission and reception traffic of a base station; a channel feature extractor configured to extract statistical features related to wireless channel conditions and wireless channel configurations; A context unit configured to generate a context vector based on statistical features related to service requirements, statistical features related to the transmission and reception traffic, and statistical features related to the radio channel conditions and radio channel configuration. A first ML model array including a plurality of machine learning (ML) models, each configured to estimate a quality of service (QoS) of a mixed numerology configuration based on the context vector generated by the context unit. An apparatus comprising a determination unit configured to select a mixed numerology configuration used for data transmission and reception based on the QoS estimated by the ML model.

[0066] [Appendix 2] The apparatus according to Appendix 1, wherein the context unit is configured to concatenate statistical features related to the service requirements, statistical features related to the overall transmission and reception traffic, and statistical features related to the radio channel conditions and the radio channel configuration to generate the context vector.

[0067] [Appendix 3] The apparatus according to Appendix 1 or 2, wherein the QoS is a ratio of the number of QoS flows corresponding to the UE for which the service requirements are satisfied to the total number of the QoS flows.

[0068] [Appendix 4] The apparatus according to any one of Appendices 1 to 3, wherein each of the ML models included in the first ML model array is pre-trained in an offline manner using collected real data or synthetic data generated by communication network simulation.

[0069] [Appendix 5] The apparatus according to any one of Appendices 1 to 4, wherein the context vector is input to each of the ML models included in the first ML model array, and the ML models output estimated QoS for different mixed numerology configurations.

[0070] [Appendix 6] A performance monitoring unit configured to calculate the QoS obtained for the mixed numerology configuration selected by the determination unit; A database configured to store the context vector, the selected mixed numerology configuration, and the QoS calculated for the selected mixed numerology configuration as context information; The apparatus according to any one of Appendices 1 to 5, further comprising an ML model update unit configured to update the ML models included in the first ML model array based on the context information.

[0071] [Appendix 7] The ML model update unit includes: A pair of a second ML model array and a third ML model array, wherein the second ML model array and the third ML model array each include a plurality of ML models configured to estimate QoS for the mixed numerology configuration based on the context vector; A control unit configured to obtain the context information from the database and update the ML models included in the first ML model array using the obtained context information, the second ML model array, and the third ML model array. The apparatus according to Appendix 6.

[0072] [Appendix 8] The control unit includes: Generating a test mini-batch and a training mini-batch of the context information; Updating the parameters of the ML models included in the third ML model array; Inputting the test mini-batch into the second ML model array and the third ML model array, and causing the second ML model array and the third ML model array to estimate QoS respectively; Cause the error of the QoS estimated by the second ML model array and the error of the QoS estimated by the third ML model array to be calculated, Based on the calculated error, determine whether to update the ML model included in the first ML model array, If it is determined to update the ML model included in the first ML model array, update the parameters of the ML model included in the first ML model array with the parameters of the ML model included in the third ML model array, The apparatus according to appendix 7, wherein the parameters of the ML model included in the third ML model array are copied to the ML model included in the second ML model array.

[0073] [Appendix 9] The requirement feature extractor, Extracts a value indicating QoS from a data packet, Obtains a service requirement corresponding to each of the extracted values using a table that maps the values to service requirements, The apparatus according to any one of appendices 1 to 8, wherein at least one statistical value of the obtained service requirements is calculated as a statistical feature related to the service requirements.

[0074] [Appendix 10] The apparatus according to any one of appendices 1 to 9, wherein the service requirements include at least one of a delay budget of a QoS flow corresponding to the UE and a packet error rate tolerance range.

[0075] [Appendix 11] The traffic feature extractor, Collects traffic feature information, Based on the collected traffic feature information, calculates at least one statistical value as a statistical feature related to the transmission and reception traffic. The apparatus according to any one of appendices 1 to 10.

[0076] [Appendix 12] The traffic feature information includes at least one of a UE ID (Identifier), a packet size, and a packet arrival time stamp, and the device according to Appendix 11.

[0077] [Appendix 13] The channel feature extractor collects channel feature information, and calculates at least one statistical value as the extracted statistical feature related to the transmission / reception traffic based on the collected channel function information, and the device according to any one of Appendices 1 to 12.

[0078] [Appendix 14] The channel feature information includes at least one of a channel quality indicator (CQI) report, the number of available resource blocks, a guard band value, a downlink and uplink share ratio, and a hybrid automatic repeat request (HARQ) process identifier, and the device according to Appendix 13.

[0079] [Appendix 15] The determination unit is configured to select a mixed numerology configuration having the highest estimated QoS among the QoSs estimated by the ML models included in the first ML model array, and the device according to any one of Appendices 1 to 14.

[0080] [Appendix 16] a communication unit configured to communicate with the core network and the UE, and a mixed numerology configuration selection unit configured to select a mixed numerology configuration used for data transmission and reception, The mixed numerology configuration selection unit includes a requirement feature extractor configured to extract statistical features related to the service requirements of the UE, a traffic feature extractor configured to extract statistical features related to the transmission / reception traffic of the base station, A channel feature extractor configured to extract statistical features related to wireless channel conditions and wireless channel configurations, A context unit configured to generate a context vector based on statistical features related to the service requirements, statistical features related to the transmission and reception traffic, and statistical features related to the wireless channel conditions and wireless channel configurations, A first ML model array including a plurality of machine learning (ML) models each configured to estimate the quality of service (QoS) of the mixed numerology configuration based on the context vector generated by the context unit, A base station having a determination unit configured to select a mixed numerology configuration used for data transmission and reception based on the QoS estimated by the ML model.

[0081] [Appendix 17] The base station according to Appendix 16, wherein the context unit is configured to concatenate statistical features related to the service requirements, statistical features related to overall transmission and reception traffic, and statistical features related to the wireless channel conditions and the wireless channel configurations to generate the context vector.

[0082] [Appendix 18] The base station according to Appendix 16 or 17, wherein the QoS is characterized by a ratio of the number of QoS flows corresponding to the UE for which the service requirements are satisfied to the total number of the QoS flows.

[0083] [Appendix 19] A method for mixed numerology configuration selection, comprising: extracting statistical features related to service requirements of a UE related to a base station; extracting statistical features related to transmission and reception traffic of the base station; extracting statistical features related to wireless channel conditions and wireless channel configurations; Generate a context vector based on statistical features related to service requirements, statistical features related to transmission and reception traffic, and statistical features related to radio channel conditions and radio channel configurations. Use a machine learning (ML) model array including a plurality of ML models each configured to estimate QoS of a mixed numerology configuration, and estimate the quality of service (QoS) of the mixed numerology configuration based on the context vector. A method comprising selecting a mixed numerology configuration used for data transmission and data reception based on the QoS estimated by the ML model.

[0084] [Appendix 20] Extract statistical features related to the service requirements of UEs related to the base station. Extract statistical features related to the transmission and reception traffic of the base station. Extract statistical features related to radio channel conditions and radio channel configurations. Generate a context vector based on statistical features related to service requirements, statistical features related to transmission and reception traffic, and statistical features related to radio channel conditions and radio channel configurations. Use a machine learning (ML) model array including a plurality of ML models each configured to estimate QoS of a mixed numerology configuration, and estimate the quality of service (QoS) of the mixed numerology configuration based on the context vector. A computer-readable medium for storing a program for causing a computer to select a mixed numerology configuration used for data transmission and data reception based on the QoS estimated by the ML model.

Explanation of Signs

[0085] 10: Base station 15: Communication unit 20: Hybrid Numerology Configuration Selection Unit 21: Requirement Feature Extractor 22: Traffic Feature Extractor 23: Channel Feature Extractor 24: Context Unit 25: ML Model Array 26: Decision Unit 30: ML Model 100: Communication System 110: 5G Core Network 120: UE 300: Base Station (gNB) 302: Hybrid Numerology Selection Unit (NSU) 303: Storage Buffer 304A: Requirement Feature Extractor 304B: Traffic Feature Extractor 304C: Channel Feature Extractor 305: Context Unit 306: ML Model Array 306-1 to 306-M: ML Model 307: Decision Unit 308: Performance Monitoring Unit 309: Database 310: 5QI Lookup Table 321: SDAP Layer Module 322: PDCP Layer Module 323: RLC Layer Module 324: MAC Layer Module 325: PHY Layer Module 326: Antenna Array 601: O-RAN Radio Unit (O-RU) 602: O-RAN Distributed Unit (O-DU) 603: O-RAN Central Unit (O-CU) 650: ML Model Update Unit 651: Control Unit 652: Context Buffer 653, 654: ML Model Array

Claims

1. An apparatus for hybrid numerology configuration selection, comprising: a requirement feature extractor configured to extract statistical features related to service requirements of UEs associated with a base station; a traffic feature extractor configured to extract statistical features related to transmission and reception traffic of the base station; a channel feature extractor configured to extract statistical features related to radio channel conditions and radio channel configurations; a context unit configured to generate a context vector based on statistical features related to service requirements, statistical features related to the transmission and reception traffic, and statistical features related to the radio channel conditions and radio channel configurations; a first ML model array including a plurality of machine learning (ML) models, each configured to estimate a quality of service (QoS) of a hybrid numerology configuration based on the context vector generated by the context unit; and a decision unit configured to select a hybrid numerology configuration to be used for data transmission and reception based on the QoS estimated by the ML model.

2. The apparatus according to claim 1, wherein the context unit is configured to concatenate statistical features related to the service requirements, statistical features related to overall transmission and reception traffic, and statistical features related to the radio channel conditions and the radio channel configurations to generate the context vector.

3. The apparatus according to claim 1 or 2, wherein the QoS is a ratio of the number of QoS flows corresponding to the UEs for which the service requirements are satisfied to the total number of the QoS flows.

4. The apparatus according to any one of claims 1 to 3, wherein the context vector is input to each of the ML models included in the first ML model array, and the ML models output estimated QoSs for different hybrid numerology configurations.

5. a performance monitoring unit configured to calculate the QoS obtained for the hybrid numerology configuration selected by the decision unit; and a database configured to store the context vector, the selected hybrid numerology configuration, and the QoS calculated for the selected hybrid numerology configuration as context information. An apparatus according to any one of claims 1 to 4, further comprising an ML model update unit configured to update an ML model included in a first ML model array based on the context information.

6. The ML model update unit A pair of a second ML model array and a third ML model array, wherein the second ML model array and the third ML model array each include a plurality of ML models configured to estimate QoS for the hybrid numerology configuration based on the context vector, the second ML model array and the third ML model array; An apparatus according to claim 5, comprising: a control unit configured to acquire the context information from the database and update an ML model included in the first ML model array using the acquired context information, the second ML model array, and the third ML model array.

7. The control unit Generates a test mini-batch and a training mini-batch of the context information, Updates parameters of the ML models included in the third ML model array, Inputs the test mini-batch to the second ML model array and the third ML model array, causes the second ML model array and the third ML model array to estimate QoS respectively, Calculates an error in QoS estimated by the second ML model array and an error in QoS estimated by the third ML model array, Determines whether to update the ML model included in the first ML model array based on the calculated error, If it is determined to update the ML model included in the first ML model array, updates the parameters of the ML model included in the first ML model array with the parameters of the ML model included in the third ML model array, Copies the parameters of the ML model included in the third ML model array to the ML model included in the second ML model array. The apparatus according to claim 6.

8. A communication unit configured to communicate with a core network and a UE, A hybrid numerology configuration selection unit configured to select a hybrid numerology configuration used for data transmission and reception, The hybrid numerology configuration selection unit A requirement feature extractor configured to extract statistical features related to service requirements of the UE A traffic feature extractor configured to extract statistical features related to the transmission and reception traffic of a base station, A channel feature extractor configured to extract statistical features related to radio channel conditions and radio channel configurations, A context unit configured to generate a context vector based on the statistical features related to the service requirements, the statistical features related to the transmission and reception traffic, and the statistical features related to the radio channel conditions and radio channel configurations, A first ML model array including a plurality of machine learning (ML) models each configured to estimate the quality of service (QoS) of the mixed numerology configuration based on the context vector generated by the context unit, A base station having a determination unit configured to select a mixed numerology configuration used for data transmission and reception based on the QoS estimated by the ML model.

9. A method for mixed numerology configuration selection, A computer extracts statistical features related to the service requirements of UEs related to a base station, The computer extracts statistical features related to the transmission and reception traffic of the base station, The computer extracts statistical features related to radio channel conditions and radio channel configurations, The computer generates a context vector based on the statistical features related to the service requirements, the statistical features related to the transmission and reception traffic, and the statistical features related to the radio channel conditions and radio channel configurations, The computer uses a machine learning (ML) model array including a plurality of ML models each configured to estimate the QoS of the mixed numerology configuration, and estimates the quality of service (QoS) of the mixed numerology configuration based on the context vector, The method includes the computer selecting a mixed numerology configuration used for data transmission and data reception based on the QoS estimated by the ML model.

10. Extract statistical features related to the service requirements of UEs related to a base station, Extract statistical features related to the transmission and reception traffic of the base station, Extract statistical features related to wireless channel conditions and wireless channel configurations, generate a context vector based on statistical features related to service requirements, statistical features related to transmission and reception traffic, and statistical features related to wireless channel conditions and wireless channel configurations, use a machine learning (ML) model array including a plurality of ML models each configured to estimate the QoS of a mixed numerology configuration, and estimate the quality of service (QoS) of the mixed numerology configuration based on the context vector, A program for causing a computer to select a mixed numerology configuration used for data transmission and data reception based on the QoS estimated by the ML model.

Images